The different particle pollution degree in transformer oil samples were made up, the infrared spectrum data of the oil samples were acquired by using the infrared spectrum scanning. Using the successive projections algorithm, the effective wavelength variables of the oil samples were extracted. Based on the extracted wavelength variables, two models of both the effective wavelength of the infrared spectrum and the particle contamination pollution degree were established by using partial least squares and support vector machine method. The effective wavelength of successive projections algorithm extracted from the infrared spectrum of transformer oil samples has the characteristics of the wavelength of specific particle contamination, and the prediction effects of the models are better than partial least squares model and support vector machine model using the full infrared spectrum data of the oil samples. Besides, the determination coefficient of the prediction set of oil samples are 0.892 9, 0.934 3 respectively with the two models, and the root mean square error are 6.372×10－3、3.07×10－3 respectively, thereby the satisfactory prediction results has achieved, these provide a reference for the detection of the particle contamination in transformer oil.